## archive
Blog
4 articles · 2 categories · type to filter
★ featured
- essays/ code-evolve: PI agent swarms code-evolve uses a swarm of coding agents and a scoring script to hill climb a program, one measured change at a time.
- notes/ Stop Asking the LLM to Remember the Rules A binary decision tree lets code ask an LLM one narrow question at a time, making large-scale classification more reliable and mistakes easier to inspect.
- essays/ Why Your AI Coding Agents Fail After 20 Minutes (And How to Fix It) Loop engineering doesn't scale past context degradation. Isolated subagents, a boring DAG harness, and shell checks as the judge are what make 8-hour unsupervised runs land.
- essays/ Spec-Driven AI Development An experiment: treat requirements as the source of truth, regenerate code when they change, and let mutation testing guard the architecture — so I ship across ~20 apps without reading the diffs.
no matches ·